Struct FastEmbedReranker 

pub struct FastEmbedReranker { /* private fields */ }

FastEmbed-backed cross-encoder reranker using ms-marco-MiniLM-L-6-v2.

This reranker uses a cross-encoder model that processes query-document pairs simultaneously, providing more accurate relevance scores than bi-encoder similarity. The MiniLM-L-6-v2 model is optimized for fast inference on CPU.

Implementations

impl FastEmbedReranker

pub fn reranker_id_static() -> &'static str

Stable reranker identifier for ms-marco-MiniLM-L-6-v2.

pub fn model_id_static() -> &'static str

Stable model identifier for ms-marco-MiniLM-L-6-v2.

pub fn required_model_files() -> &'static [&'static str]

Required model files for the reranker (must all exist locally).

pub fn default_model_dir(data_dir: &Path) -> PathBuf

Default model directory relative to a data directory.

pub fn load_from_dir(model_dir: &Path) -> Result<FastEmbedReranker, SearchError>

Load the reranker model + tokenizer from a local directory.

This never downloads; it returns an error if any required file is missing.

pub fn model_id(&self) -> &str

Stable model identifier for compatibility checks.

Trait Implementations

impl SyncRerank for FastEmbedReranker

fn rerank_sync( &self, query: &str, documents: &[RerankDocument], ) -> Result<Vec<RerankScore>, SearchError>

Synchronously rerank documents against a query.

fn id(&self) -> &str

A unique identifier for this reranker model.

fn model_name(&self) -> &str

Human-friendly reranker model name.

fn is_available(&self) -> bool

Whether this reranker is loaded and ready for inference.

fn max_length(&self) -> usize

Maximum supported token length for query+document pair input.